Skip to content
CAI
Software that uses CAICheck a score

resemble-ai/chatterbox

64.6

Adequate · 18 September 2026

8k

lines of production code

Python

primary language

1

measurement over time

CAI band scale
CAI lens gauges

What this system is

Chatterbox is a multilingual text-to-speech system that synthesizes speech in over 20 languages using the S3Gen meanflow architecture. It supports various model backends, including Llama and GPT-2 variants, to offer different performance and size options such as Turbo and Nano. The system provides Gradio-based interfaces and example scripts for inference, with updated dependencies ensuring compatibility with modern Python versions.

Features

Chatterbox introduces multilingual TTS and S3Gen meanflow architecture

Chatterbox now supports multilingual text-to-speech synthesis via the new ChatterboxMultilingualTTS class, which handles 22 languages and allows users to opt into a v3 checkpoint. The underlying audio generation engine has been upgraded to S3Gen, introducing a meanflow mode that modifies the decoder and flow-matching inference to improve generation quality. Additionally, the T3 model backend now supports GPT-2 architectures (Medium and Small) alongside Llama, enabling smaller model variants like Chatterbox Nano.

src/chatterbox · high confidence

New Turbo and Multilingual V3 demos, updated example scripts, and removed legacy voice conversion tool

This update introduces new Gradio applications and example scripts for the Chatterbox-Turbo and Chatterbox-Nano models, including a Turbo-specific UI with paralinguistic tag support and a multilingual V3 app with 23+ language defaults. The standard TTS example scripts (example\_tts.py, example\_vc.py) and the main Gradio app (gradio\_tts\_app.py) have been updated to support automatic device detection (including MPS for Mac) and new sampling parameters (min\_p, top\_p, repetition\_penalty). The legacy voice\_conversion.py script has been removed, and the README has been updated to reflect the new model zoo and documentation.

(repo-wide) · high confidence

Dependencies

Update dependencies for Python 3.13+ compatibility and new features

The project now requires Python 3.10 or higher and updates core dependencies to support newer Python versions, including conditional version ranges for NumPy (1.x for \<3.13, 2.x for 3.13+) and PyTorch/Torchaudio (2.6.0 for \<3.14, 2.9.0+ for 3.14+). Librosa is upgraded to 0.11.0, and the Transformers library is updated to version 5.2.0. New dependencies added include Gradio 6.8.0, safetensors, spacy-pkuseg, pykakasi, and pyloudnorm, while resemble-perth is now installed directly from its Git repository.

(dependencies) · high confidence

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

How this codebase got here

Baseline

  • First survey — no prior run to compare against. CAI 65.

Lenses

  • Code Health 99
  • Architecture 100
  • Maturity 63
  • Readiness 47
  • Security 97

Changes since last survey

  • 42 commits — 37 feature/other, 5 fixes

By area

  • (root) — 28 commits
  • src/chatterbox — 10 commits
  • (repo) — 4 commits

Notable commits

  • fix: Fix ReadMe Banner Image (#381)
  • fix: Fix link in readme
  • fix: Potentially fix CUDA error (#114)
  • fix: Update README.md with fixed image link (#236)
  • fix: fix: add map_location to torch.load for CPU/MPS support in multilingual model (#410)
  • change: Add Podonos evaluation section and acknowledgement to README (#456)
  • change: Add opt-in v3 multilingual checkpoint, skip analyzer for v3 (#516)
  • change: Apply CFG optionally based on cfg_weight
  • change: Broaden dependency version ranges for Python 3.13+ compatibility (#495)
  • change: Chatterbox Turbo 350M Model (#380)
  • change: Clarify OS & Py version; Adjust dependencies (#138)
  • change: Create example_for_mac.py
  • change: Enable usage of Min_P Sampler, modify other sampler settings (#155)
  • change: Feature: Add cpu mps support (#35)
  • change: Feature: Update example scripts, add example wavs, add info on watermarking, safetensor in VC model (#82)
  • change: Make russian-text-stresser optional (#376)
  • change: Merge pull request #40 from serome111/master
  • change: Merge pull request #75 from resemble-ai/disc
  • change: Merge pull request #80 from resemble-ai/optinal_cfg
  • change: Merge pull request #81 from resemble-ai/update_safetensor
  • …and 22 more

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

Survey your own repository

resemble-ai/chatterbox was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.

About this page

  • The score is its most recent published measurement, taken on 18 September 2026 at a pinned commit. It is not a live figure and does not change until the project is measured again.
  • Measured at commit 5de7a54aa4e5e2baadb0182dde554908b48b85c2 — the exact code this score is about.
  • Scored under rubric-2026.09.15 — the same rubric and the same method as every other entry in this index.
  • Measured by watchdog.canine.dev using codehealth-analyzer preprod-5d04157a340d.